An AI Deployment Framework

October 30, 2020

AI Deployment Framework
Interests in AI, and particularly AI for health care has grown immensely over the recent years.  However, enthusiasm for AI in healthcare has been overshadowed by the challenging path to successful implementation of these solutions into routine clinical care.  Many of these challenges stem from the 

1. lack of buy-in, from necessary stakeholders, 
2. lack of evidence for clinical utility, and 
3. lack of resources and expertise in AI deployment. 

In order to help implement an AI solution into the healthcare setting, its helpful to think about a holistic framework for translation of the product into the delivery system.  The FDA offers such a framework – the total product lifecycle or TPLC of an AI solution. 

The 4 Components of AI Deployment
Design and Development

• Identifying the right problem to solve
• Evaluate whether that problem can be solved (or is worth solving) using AI.
• Understand the type of data do you need to solve this problem

Evaluation and validation

• Includes multiple iterations.
• Evaluate the model using retrospective data.
• Evaluation of net utility and work capacity.
• Prospective clinical validation studies (silent mode)

Diffuse and scale

• Understand the role of interoperability
• Know the importance of multi-institutional deployment
• Describe the role of different funders

Continuous monitoring and maintenance

• Continuously learning from new data
• Evaluating safety and efficiency once deployed
• Bug fixes and maintenance
• Incorporating new data and evidence

Source: https://www.coursera.org/learn/evaluations-ai-applications-healthcare/